Artificial Intelligence-Driven Antenna Beamforming for Multi-User MIMO in Satellite Broadband Systems

Staffieri A.; Staffieri F.; Staffieri C.; Gupta M.; Bett J.S.P.; Alzadjali B.; Bhagat A.; Fathy N. · Proceedings of the International Astronautical Congress, IAC (2025) · DOI: 10.52202/083082-0069

Resumen

This research proposes an AI-powered algorithm for multi-user multi-input multi-output (MIMO) satellite broadband communications for antenna beamforming. With growing requirements for high-datarate and low-latency satellite networks, new approaches must effectively counteract channel dynamics, user mobility, and interference suppression. Traditional approaches to beamforming often face difficulty in effectively countering these variables, with a strong reliance on predefined algorithms and a lack of real-time capabilities for optimization. We present a deep learning-powered model employing recurrent neural networks (RNNs) and reinforcement learning (RL) for real-time dynamic optimization of beam coefficients and direction of steering. The AI model is trained with rich datasets derived from real-life propagation scenarios, including non-stationary distributions of users, atmospheric factors, and intersatellite interference. With such an approach, the system can make future-optimum prediction of beam configuration, minimizing interference during transmission and improving spectral efficiency. The proposed scheme incorporates a hybrid precoding mechanism combining digital and analog techniques for beamforming, minimizing both hardware complexity and power consumption, beneficial for satellite platforms with restricted capabilities. Simulative studies reveal improvements in terms of 35% in terms of spectral efficiency and 20% in terms of reduced bit error rate (BER) over conventional approaches for a range of loading and mobility scenarios. In addition, the proposed scheme proves to exhibit high robustness in dealing with non-line-of-sight (NLoS) scenarios and multipath fading, providing reliable service delivery over a range of geographical locations. The principal purpose of this work is to develop a new algorithm for tracking a beam with a use case for mobile users, in addition to utilizing federated learning for supporting distributed training at ground terminals with a safeguard for protecting data privacy. Together, these breakthroughs lay a basis for developing future, flexible satellite broadband networks for supporting new emerging applications, such as IoT connectivity, remote sensing, and ubiquitous access to the internet. Following studies will seek to extend the model to include heterogeneous network structures and explore meta-learning approaches for enhancing adaptability in changing environments relevant to communications in space. This work identifies artificial intelligence’s transformational role in modern satellite MIMO technology for future-generation broadband service.

Instituciones: Universidad Nacional de Ingeniería